Assessing brain involvement in Fabry disease with deep learning and the brain-age paradigm.

While neurological manifestations are core features of Fabry disease (FD), quantitative neuroimaging biomarkers allowing to measure brain involvement are lacking. We used deep learning and the brain-age paradigm to assess whether FD patients' brains appear older than normal and to validate brain-predicted age difference (brain-PAD) as a possible disease severity biomarker. MRI scans of FD patients and healthy controls (HCs) from a single Institution were, retrospectively, studied. The Fabry stabilization index (FASTEX) was recorded as a measure of disease severity. Using minimally preprocessed 3D T1-weighted brain scans of healthy subjects from eight publicly available sources (N = 2160; mean age = 33 years [range 4-86]), we trained a model predicting chronological age based on a DenseNet architecture and used it to generate brain-age predictions in the internal cohort. Within a linear modeling framework, brain-PAD was tested for age/sex-adjusted associations with diagnostic group (FD vs. HC), FASTEX score, and both global and voxel-level neuroimaging measures. We studied 52 FD patients (40.6 ± 12.6 years; 28F) and 58 HC (38.4 ± 13.4 years; 28F). The brain-age model achieved accurate out-of-sample performance (mean absolute error = 4.01 years, R2 = .90). FD patients had significantly higher brain-PAD than HC (estimated marginal means: 3.1 vs. -0.1, p = .01). Brain-PAD was associated with FASTEX score (B = 0.10, p = .02), brain parenchymal fraction (B = -153.50, p = .001), white matter hyperintensities load (B = 0.85, p = .01), and tissue volume reduction throughout the brain. We demonstrated that FD patients' brains appear older than normal. Brain-PAD correlates with FD-related multi-organ damage and is influenced by both global brain volume and white matter hyperintensities, offering a comprehensive biomarker of (neurological) disease severity.

VeerNet: Using Deep Neural Networks for Curve Classification and Digitization of Raster Well-Log Images.

Raster logs are scanned representations of the analog data recorded in subsurface drilling. Geologists rely on these images to interpret well-log curves and deduce the physical properties of geological formations. Scanned images contain various artifacts, including hand-written texts, brightness variability, scan defects, etc. The manual effort involved in reading the data is substantial. To mitigate this, unsupervised computer vision techniques are employed to extract and interpret the curves digitally. Existing algorithms predominantly require manual intervention, resulting in slow processing times, and are erroneous. This research aims to address these challenges by proposing VeerNet, a deep neural network architecture designed to semantically segment the raster images from the background grid to classify and digitize (i.e., extracting the analytic formulation of the written curve) the well-log data. The proposed approach is based on a modified UNet-inspired architecture leveraging an attention-augmented read-process-write strategy to balance retaining key signals while dealing with the different input-output sizes. The reported results show that the proposed architecture efficiently classifies and digitizes the curves with an overall F1 score of 35% and Intersection over Union of 30%, achieving 97% recall and 0.11 Mean Absolute Error when compared with real data on binary segmentation of multiple curves. Finally, we analyzed VeerNet's ability in predicting Gamma-ray values, achieving a Pearson coefficient score of 0.62 when compared to measured data.

ReUse: REgressive Unet for Carbon Storage and Above-Ground Biomass Estimation.

The United Nations Framework Convention on Climate Change (UNFCCC) has recently established the Reducing Emissions from Deforestation and forest Degradation (REDD+) program, which requires countries to report their carbon emissions and sink estimates through national greenhouse gas inventories (NGHGI). Thus, developing automatic systems capable of estimating the carbon absorbed by forests without in situ observation becomes essential. To support this critical need, in this work, we introduce ReUse, a simple but effective deep learning approach to estimate the carbon absorbed by forest areas based on remote sensing. The proposed method's novelty is in using the public above-ground biomass (AGB) data from the European Space Agency's Climate Change Initiative Biomass project as ground truth to estimate the carbon sequestration capacity of any portion of land on Earth using Sentinel-2 images and a pixel-wise regressive UNet. The approach has been compared with two literature proposals using a private dataset and human-engineered features. The results show a more remarkable generalization ability of the proposed approach, with a decrease in Mean Absolute Error and Root Mean Square Error over the runner-up of 16.9 and 14.3 in the area of Vietnam, 4.7 and 5.1 in the area of Myanmar, 8.0 and 1.4 in the area of Central Europe, respectively. As a case study, we also report an analysis made for the Astroni area, a World Wildlife Fund (WWF) natural reserve struck by a large fire, producing predictions consistent with values found by experts in the field after in situ investigations. These results further support the use of such an approach for the early detection of AGB variations in urban and rural areas.

A Multimodal Knowledge-Based Deep Learning Approach for MGMT Promoter Methylation Identification.

Glioblastoma Multiforme (GBM) is considered one of the most aggressive malignant tumors, characterized by a tremendously low survival rate. Despite alkylating chemotherapy being typically adopted to fight this tumor, it is known that O(6)-methylguanine-DNA methyltransferase (MGMT) enzyme repair abilities can antagonize the cytotoxic effects of alkylating agents, strongly limiting tumor cell destruction. However, it has been observed that MGMT promoter regions may be subject to methylation, a biological process preventing MGMT enzymes from removing the alkyl agents. As a consequence, the presence of the methylation process in GBM patients can be considered a predictive biomarker of response to therapy and a prognosis factor. Unfortunately, identifying signs of methylation is a non-trivial matter, often requiring expensive, time-consuming, and invasive procedures. In this work, we propose to face MGMT promoter methylation identification analyzing Magnetic Resonance Imaging (MRI) data using a Deep Learning (DL) based approach. In particular, we propose a Convolutional Neural Network (CNN) operating on suspicious regions on the FLAIR series, pre-selected through an unsupervised Knowledge-Based filter leveraging both FLAIR and T1-weighted series. The experiments, run on two different publicly available datasets, show that the proposed approach can obtain results comparable to (and in some cases better than) the considered competitor approach while consisting of less than 0.29% of its parameters. Finally, we perform an eXplainable AI (XAI) analysis to take a little step further toward the clinical usability of a DL-based approach for MGMT promoter detection in brain MRI.

Comparison between two packages for pectoral muscle removal on mammographic images.

BACKGROUND: Pectoral muscle removal is a fundamental preliminary step in computer-aided diagnosis systems for full-field digital mammography (FFDM). Currently, two open-source publicly available packages (LIBRA and OpenBreast) provide algorithms for pectoral muscle removal within Matlab environment. PURPOSE: To compare performance of the two packages on a single database of FFDM images. METHODS: Only mediolateral oblique (MLO) FFDM was considered because of large presence of pectoral muscle on this type of projection. For obtaining ground truth, pectoral muscle has been manually segmented by two radiologists in consensus. Both LIBRA's and OpenBreast's removal performance with respect to ground truth were compared using Dice similarity coefficient and Cohen-kappa reliability coefficient; Wilcoxon signed-rank test has been used for assessing differences in performances; Kruskal-Wallis test has been used to verify possible dependence of the performance from the breast density or image laterality. RESULTS: FFDMs from 168 consecutive women at our institution have been included in the study. Both LIBRA's Dice-index and Cohen-kappa were significantly higher than OpenBreast (Wilcoxon signed-rank test P < 0.05). No dependence on breast density or laterality has been found (Kruskal-Wallis test P > 0.05). CONCLUSION: Libra has a better performance than OpenBreast in pectoral muscle delineation so that, although our study has not a direct clinical application, these results are useful in the choice of packages for the development of complex systems for computer-aided breast evaluation.

Stratification of multiple sclerosis patients using unsupervised machine learning: a single-visit MRI-driven approach.

OBJECTIVES: To stratify patients with multiple sclerosis (pwMS) based on brain MRI-derived volumetric features using unsupervised machine learning. METHODS: The 3-T brain MRIs of relapsing-remitting pwMS including 3D-T1w and FLAIR-T2w sequences were retrospectively collected, along with Expanded Disability Status Scale (EDSS) scores and long-term (10 ± 2 years) clinical outcomes (EDSS, cognition, and progressive course). From the MRIs, volumes of demyelinating lesions and 116 atlas-defined gray matter regions were automatically segmented and expressed as z-scores referenced to external populations. Following feature selection, baseline MRI-derived biomarkers entered the Subtype and Stage Inference (SuStaIn) algorithm, which estimates subgroups characterized by distinct patterns of biomarker evolution and stages within subgroups. The trained model was then applied to longitudinal MRIs. Stability of subtypes and stage change over time were assessed via Krippendorf's α and multilevel linear regression models, respectively. The prognostic relevance of SuStaIn classification was assessed with ordinal/logistic regression analyses. RESULTS: We selected 425 pwMS (35.9 ± 9.9 years; F/M: 301/124), corresponding to 1129 MRI scans, along with healthy controls (N = 148; 35.9 ± 13.0 years; F/M: 77/71) and external pwMS (N = 80; 40.4 ± 11.9 years; F/M: 56/24) as reference populations. Based on 11 biomarkers surviving feature selection, two subtypes were identified, designated as "deep gray matter (DGM)-first" subtype (N = 238) and "cortex-first" subtype (N = 187) according to the atrophy pattern. Subtypes were consistent over time (α = 0.806), with significant annual stage increase (b = 0.20; p < 0.001). EDSS was associated with stage and DGM-first subtype (p ≤ 0.02). Baseline stage predicted long-term disability, transition to progressive course, and cognitive impairment (p ≤ 0.03), with the latter also associated with DGM-first subtype (p = 0.005). CONCLUSIONS: Unsupervised learning modelling of brain MRI-derived volumetric features provides a biologically reliable and prognostically meaningful stratification of pwMS. KEY POINTS: • The unsupervised modelling of brain MRI-derived volumetric features can provide a single-visit stratification of multiple sclerosis patients. • The so-obtained classification tends to be consistent over time and captures disease-related brain damage progression, supporting the biological reliability of the model. • Baseline stratification predicts long-term clinical disability, cognition, and transition to secondary progressive course.

Interaction data are identifiable even across long periods of time.

Fine-grained records of people's interactions, both offline and online, are collected at large scale. These data contain sensitive information about whom we meet, talk to, and when. We demonstrate here how people's interaction behavior is stable over long periods of time and can be used to identify individuals in anonymous datasets. Our attack learns the profile of an individual using geometric deep learning and triplet loss optimization. In a mobile phone metadata dataset of more than 40k people, it correctly identifies 52% of individuals based on their 2-hop interaction graph. We further show that the profiles learned by our method are stable over time and that 24% of people are still identifiable after 20 weeks. Our results suggest that people with well-balanced interaction graphs are more identifiable. Applying our attack to Bluetooth close-proximity networks, we show that even 1-hop interaction graphs are enough to identify people more than 26% of the time. Our results provide strong evidence that disconnected and even re-pseudonymized interaction data can be linked together making them personal data under the European Union's General Data Protection Regulation.

Robustness of Radiomics in Pre-Surgical Computer Tomography of Non-Small-Cell Lung Cancer.

Background: Radiomic features are increasingly used in CT of NSCLC. However, their robustness with respect to segmentation variability has not yet been demonstrated. The aim of this study was to assess radiomic features agreement across three kinds of segmentation. Methods: We retrospectively included 48 patients suffering from NSCLC who underwent pre-surgery CT. Two expert radiologists in consensus manually delineated three 3D-ROIs on each patient. To assess robustness for each feature, the intra-class correlation coefficient (ICC) across segmentations was evaluated. The 'sensitivity' of ICC upon some parameters affecting features computation (such as bin-width for first-order features and pixel-distances for second-order features) was also evaluated. Moreover, an assessment with respect to interpolator and isotropic resolution was also performed. Results: Our results indicate that 'shape' features tend to have excellent agreement (ICC > 0.9) across segmentations; moreover, they have approximately zero sensitivity to other parameters. 'First-order' features are in general sensitive to parameters variation; however, a few of them showed excellent agreement and low sensitivity (below 0.1) with respect to bin-width and pixel-distance. Similarly, a few second-order features showed excellent agreement and low sensitivity. Conclusions: Our results suggest that a limited number of radiomic features can achieve a high level of reproducibility in CT of NSCLC.

A Pipelined Tracer-Aware Approach for Lesion Segmentation in Breast DCE-MRI.

The recent spread of Deep Learning (DL) in medical imaging is pushing researchers to explore its suitability for lesion segmentation in Dynamic Contrast-Enhanced Magnetic-Resonance Imaging (DCE-MRI), a complementary imaging procedure increasingly used in breast-cancer analysis. Despite some promising proposed solutions, we argue that a "naive" use of DL may have limited effectiveness as the presence of a contrast agent results in the acquisition of multimodal 4D images requiring thorough processing before training a DL model. We thus propose a pipelined approach where each stage is intended to deal with or to leverage a peculiar characteristic of breast DCE-MRI data: the use of a breast-masking pre-processing to remove non-breast tissues; the use of Three-Time-Points (3TP) slices to effectively highlight contrast agent time course; the application of a motion-correction technique to deal with patient involuntary movements; the leverage of a modified U-Net architecture tailored on the problem; and the introduction of a new "Eras/Epochs" training strategy to handle the unbalanced dataset while performing a strong data augmentation. We compared our pipelined solution against some literature works. The results show that our approach outperforms the competitors by a large margin (+9.13% over our previous solution) while also showing a higher generalization ability.

Machine Learning Applications in Head and Neck Radiation Oncology: Lessons From Open-Source Radiomics Challenges.

Radiomics leverages existing image datasets to provide non-visible data extraction via image post-processing, with the aim of identifying prognostic, and predictive imaging features at a sub-region of interest level. However, the application of radiomics is hampered by several challenges such as lack of image acquisition/analysis method standardization, impeding generalizability. As of yet, radiomics remains intriguing, but not clinically validated. We aimed to test the feasibility of a non-custom-constructed platform for disseminating existing large, standardized databases across institutions for promoting radiomics studies. Hence, University of Texas MD Anderson Cancer Center organized two public radiomics challenges in head and neck radiation oncology domain. This was done in conjunction with MICCAI 2016 satellite symposium using Kaggle-in-Class, a machine-learning and predictive analytics platform. We drew on clinical data matched to radiomics data derived from diagnostic contrast-enhanced computed tomography (CECT) images in a dataset of 315 patients with oropharyngeal cancer. Contestants were tasked to develop models for (i) classifying patients according to their human papillomavirus status, or (ii) predicting local tumor recurrence, following radiotherapy. Data were split into training, and test sets. Seventeen teams from various professional domains participated in one or both of the challenges. This review paper was based on the contestants' feedback; provided by 8 contestants only (47%). Six contestants (75%) incorporated extracted radiomics features into their predictive model building, either alone (n = 5; 62.5%), as was the case with the winner of the "HPV" challenge, or in conjunction with matched clinical attributes (n = 2; 25%). Only 23% of contestants, notably, including the winner of the "local recurrence" challenge, built their model relying solely on clinical data. In addition to the value of the integration of machine learning into clinical decision-making, our experience sheds light on challenges in sharing and directing existing datasets toward clinical applications of radiomics, including hyper-dimensionality of the clinical/imaging data attributes. Our experience may help guide researchers to create a framework for sharing and reuse of already published data that we believe will ultimately accelerate the pace of clinical applications of radiomics; both in challenge or clinical settings.